In my talk, Cultural Compliance and Quantifying Behavior,
I discuss ways in which Natural Language Processing and Psychometric Analysis can potentially enhance the detection of bad actors
4. How important is it to identify bad actors?
Over $204 billion has been lost since 2009
• Company Total settlments Sums paid ($billions)
• Bank of America 34 $77.09
• JPMorgan Chase 26 $40.12
• Citigroup 18 $18.39
• Wells Fargo 10 $10.24
• BNP Paribas 1 $8.90
• UBS 8 $6.54
• Deutsche Bank 4 $5.53
• Morgan Stanley 7 $4.78
• Barclays 7 $4.23
• Credit Suisse 4 $3.74
• Source: Keffe, Bruyette & Woods (settlements exceeding $100 million)
Harry.mendell@gmail.com 917-609-5179
6. • a sure {bet}|{thing}
• adjust your account|losses|profits
• against my expressed[?] wishes
• answer {your}|{the} %ANY%[0,3] phone
• are {not responsive}|unresponsive
• bad to worse
• charge in excessive amount
• charged {too much}|{excessively}
• close|end|terminate my %ANY% [1,5] relationship with
GS|Goldman|{this firm}
• Clowns {managing|running} the
fund|show|portfolio|account|{my money}
• concern* %ANY% [1,5] safety of my money|fund|account
• cover {your}|{our} losses
• didn't authorize the sale
• didn't|didnt|{did not} explain to me|us
• disturbs|troubles me|us
• don't worry i'll take care of it
• don't you f*cking understand
• done|{did this} without %ANY% [1,5]
calling|emailing|contacting me|us|anyone
• embezzled the account
• extremely|really|quite|very unhappy|disappointed
Harry.mendell@gmail.com 917-609-5179
• failed to execute {our}|{my}instructions
• fix the {trade}|{trades}|{commissions}
• fix|adjust|change the trade*|commission*
• formally|formal complain|complaint
• found numerous|several errors|mistakes
• give you a piece of {the}|{my} commission
• how could this happen again[?]
• How could you|GS|Goldman possibly[?] lose so|this|that
much
• I %ANY% [0,4] {losing}|{lost} patience with
{you}|{GS}|{Goldman}
• I am not a happy camper
• I didn't {authorize}|{agree}
• I expect {a|an|your}[?] {answer*|response}
{today|now|asap}
• I|we have lost|{run out of}|{ran out of}
confidence|faith|trust|patience
• I have raised %ANY% [1,5] at least %ANY%[1,3] times
• I have raised %ANY% [1,5] so many times
• I lost {exorbitant|enormous amounts of}|{so|too much}
money
• I told you %ANY% {days|weeks|months} ago
• I told you %ANY%[0,1] {days|weeks|months} ago
• i want the %ANY%[0,2] trade reversed
• Paying fees {through|thru} the {nose|a--|butt}
• phone {calls}|{call}e-mail{have}|{has} not been answered
• piece of sh*t
• pissed|pisses me off
• poor|terrible|crappy {fund|account|portfolio}[?]
results|performance
• really %ANY%[0,2] pissed|{PO'd}
• rebate|refund my|your loss*
• rebate|refund what I lost
• register that|this as a complaint
• remedy the situation
• report the matter to the {sec}|{nasd}|{nyse}
• reverse the commissions
• reverse {this}|{the} %ANY%[0,5] {trade}|{transaction}
• screw*|f*ck* it up
• so frustrat*
• something {went}|{is really}|{will go} wrong
• stock will {fly}|{soar}|{dive}|{tank}
• supposed to be the top|best financial company
• surprised|concerned|frustrated|angry that you didn't|{did
not} contact|call|email me
• take care of any fees|commissions
One third of Goldman’s search phrases
14. Word Vectors (Richard Socher CS244 Stanford 4/1/15)
Problems with simple co-occurrence vectors
• Increase in size with vocabulary
• Very high dimensional: require a lot of storage
• Subsequent classification models have sparsity issues
• Models are less robust
• Solution: Low dimensional vectors
• Idea: store “most” of the important information in a small, dense vector
• Usually around 25-1000 dimensions
• One solution is to take Singular Value Decomposition of the coocurrence matrix
• But computational costs scale quadratically
Harry.mendell@gmail.com 917-609-5179